In [3]:
# Reproject trapezoidal tray to rectangular (rectify)
def snap(g1, g2, threshold=1e6):
coordinates = []
for x, y in g1.coords: # for each vertex in the first line
point = Point(x, y)
p1, p2 = nearest_points(point, g2) # find the nearest point on the second line
if p1.distance(p2) <= threshold:
# it's within the snapping tolerance, use the snapped vertex
coordinates.append(p2.coords[0])
else:
# it's too far, use the original vertex
coordinates.append((x, y))
# convert coordinates back to a LineString and return
return LineString(coordinates)
img = cv2.cvtColor(cv2.imread('EX4_ID/D1/IMG_7499.JPEG'), cv2.COLOR_BGR2RGB)
masks = SamAutomaticMaskGenerator(
model=sam,
points_per_side=32, # The number of points to be sampled along one side of the image
min_mask_region_area=100, # If >0, postprocessing will be applied to remove disconnected regions and holes in masks with area smaller than min_mask_region_area
).generate(img)
tray = max(masks, key=lambda x: x["area"])
# Convert binary mask to polygon
tray = gpd.GeoSeries(shape(s) for s, v in shapes(tray["segmentation"].astype(np.uint8), mask=tray["segmentation"])).unary_union
plt.imshow(img)
trapezoid = snap(tray.envelope.exterior, tray)
bounds = trapezoid.envelope.exterior
gpd.GeoSeries(trapezoid).plot(ax=plt.gca(), color="red")
gpd.GeoSeries(bounds).plot(ax=plt.gca(), color="green")
plt.show()
source_corners = np.float32(trapezoid.coords[:4])
target_corners = np.float32(bounds.coords[:4])
matrix = cv2.getPerspectiveTransform(source_corners, target_corners)
print(matrix)
img = cv2.warpPerspective(img, matrix, (img.shape[1], img.shape[0]), cv2.INTER_LINEAR, borderMode=cv2.BORDER_CONSTANT, borderValue=(0,0,0))
plt.imshow(img)
✔️ 22.3 s (2023-10-04T16:35:55/2023-10-04T16:36:18)
[[ 8.52514856e-01 -3.31867401e-02 4.21931441e+01] [-6.01653962e-02 9.10413456e-01 2.97571369e+01] [-8.18716445e-05 -2.95251184e-05 1.00000000e+00]]
Out[3]:
<matplotlib.image.AxesImage at 0x7f09780b3580>
